The family of the late Allan Scarff has generously made publicly available a number of files of Allan's work in the area of Go-playing software and Artificial Intelligence, in the hope that others interested in these fields will find them useful.
Allan was a pioneer of computer Go, writing a number of award-winning programs for BBC Acorn microcomputers in the 1980s (see his obituary and in British Go Journal 159, Spring 2012 for details) and a highly successful program for a Nintendo Games Machine.
The files here are:
An 'e-book' consisting of a collection of html files, which Allan called Global Connectivity Strategy.
The strategy has only a few simple rules: it doesn’t tell players where to move, but it tells them what to do. The basic idea is to connect stones efficiently, which is the Prime Directive.
This may seem mundane, but it means doing nothing else (and all competing strategies do something else). The Global part of the name refers to the attention paid to frameworks. The guidelines enable a Go player to play consistent moves to good effect, no matter how strong the opponent.
In the book, Allan compares and contrasts his strategy with 11 alternative strategies, including the one he used to play himself in the early 1970s — Multiple Weak Group Strategy (British Club Style). He conceded that the general idea was not unique to him, but thought that his version was the most ordered.
A pdf file describing Allan's Artificial Intelligence system called Acolyte Artificial Neural Net System (AANNS).
Allan wrote that if a person knows how to break down the many features of the game of Go into a comprehensive set of fundamental key concepts (such as how to connect stones and how to make eyes), AANNS can, given the relevant examples, automatically synthesize these into a strong playing algorithm.
Allan was working on the specification when he was struck down with the cancer which was to terminate his life all too early. He felt it was his most ambitious and important project to date. AANNS is a generic method not limited to Go, but could potentially be used to improve Go playing software.